Obligatory "why is my generations slow" discussions.

[Axmr1]

I'll leave this one to you wizards. Last night I went the trouble of installing cudNN's and the latest versions of Cuda tookkits and what not and by golly the performance gains was amazingly dismal.

I'm rocking a 4070ti Super and Ryzen 3 3600 with 64GB ECC memory at 3200mhz.
Why is Auto1111 being a d!ck? what more does it want? am I not enough for you? you son of a ditch? was my love not enough for your worthless bottocks to feel enlightened by the glory? Does size truly matter? What gives Auto1111?

[Axmr1]

Also

Running 1024 x 1024
Eular A with "automatic" scheduler
Hi res fix being the slowest part of the process.

[Axmr1]

I guess I scared everyone off

[Haoming02]

the performance gains was amazingly dismal

Because they are never used to begin with? If you installed the Webui correctly, it already comes with the version of PyTorch that was compiled with CUDA capability. Manually installing CUDA Toolkit and cudNN is not required, and achieves nothing at all. (Unless you want to use TensorRT or other external features)

TL;DR: You wasted your own time for nothing

[Haoming02]

As for the slow speed, set CUDA - System Fallback Policy in the Nvidia Control Panel to Prefer No System Fallback